Tarun K

BatchNormsync with Adam Optimizer

5

Is the bnsync code written specifically for SGD optimizer? The loss is not converging if I use and train the model with Adam optimizer.

For a [BasicBlock](https://github.com/fyu/drn/blob/16acdba72f4115992e02a22be7e08cb3762f8e51/drn.py#L32), the dilation is d for all layers except the first; and for the first, it is d/2. What is the equivalent of this logic in the bottleneck...

Question on training scheme

1

Hi, In section 5.1, when you compare results for 8x interpolation, does this mean that the interpolation is being performed from (1000/8=)125 FPS to 1000 FPS? And what does 8x...

About Vimeo Dataset.

1

What is the frame rate for the videos in the dataset? Also, what is the sampling frequency for the triplet and septuplet datasets?

How did you shortlist videos from Kinetics? Are these motion stabilized/ less camera motion and so on? What criterion did you use?

Hi, Do you know what is the effect of domain specific batch norm on the final accuracy? What would be the drop without it?

Large-scale DA paper

1

Hi, I want to include two of my DA papers to this list: 1. [MemSAC: Memory Augmented Sample Consistency for Large Scale Domain Adaptation (ECCV'22)](https://tarun005.github.io/MemSAC/) 2. [GeoNet: Benchmarking Unsupervised Adaptation...

What is the role of entropy loss?

2

Hi, I am not able to understand the role of the entropy loss you use in your code, which was not mentioned in the paper. https://github.com/lijin118/CGDM/blob/2d8fee0c8df158fb078ceed65979abbe46282506/utils.py#L429 Is this important to...

https://github.com/lijin118/CGDM/blob/2d8fee0c8df158fb078ceed65979abbe46282506/train_domainnet.py#L78 From this line, it looks like you also use data augmentation during test-time. Are the results in the paper with this augmentation?

Pretrained model

1

what is the pretrained model for DeepLabV3 trained on? https://github.com/doiken23/DeepLab_pytorch/blob/cdee2b098c69141960fc757f04e70c6658bd6de4/DeepLab_v3.py#L9